Datasets:

Nalandadata
/

BLINDSPOT-sample

item_id stringclasses 10 values	model stringclasses 7 values	condition stringclasses 4 values	f1 float64 0.26 0.91	precision float64 0.22 1	recall float64 0.23 0.83	n_pred int64 2 15	n_key int64 2 17	predicted_names listlengths 2 15	key_names listlengths 2 17	subject stringclasses 10 values	tier stringclasses 3 values	unsolved_core bool 2 classes	recitable bool 2 classes	parse_ok bool 1 class
NALANDA-108	openai/gpt-4o-mini	no-image	0.769	0.833	0.714	6	7	[ "Variable region", "Constant region", "Hinge region", "Heavy chain", "Light chain", "Antigen-binding site" ]	[ "Antigen", "Variable region", "Light chain", "Heavy chain", "Constant region", "Antibody", "S-S (disulphide bond)" ]	Zoology diagram	core	true	true	true
NALANDA-130	openai/gpt-5.6-luna	told	0.909	1	0.833	7	6	[ "Cerebral ganglion", "Circumpharyngeal connective", "Subpharyngeal ganglion", "Segmental ganglion", "Ventral nerve cord", "Segmental nerves", "Lateral nerves" ]	[ "“Brain”", "Nerve", "Nerve cord", "Circumpharyngeal connectives", "Subpharyngeal ganglion", "Ganglion" ]	Nervous System of Earthworm	extended	true	true	true
NALANDA-133	anthropic/claude-opus-5	no-image	0.762	0.889	0.667	9	12	[ "Pineal gland", "Pituitary gland (hypophysis)", "Thyroid gland", "Parathyroid glands", "Thymus", "Adrenal gland (suprarenal gland)", "Pancreas (pancreatic islets)", "Ovary", "Testis" ]	[ "Hypothalamus", "Pituitary", "Pineal", "Thyroid", "Parathyroid", "Thymus", "Adrenal", "Kidney", "Pancreas", "Ovary", "Uterus", "Teste" ]	The main ductless glands of the human body	extended	true	true	true
NALANDA-149	openai/gpt-4o	blind	0.6	1	0.429	4	7	[ "corneal lens", "corneagenous cell", "crystalline cone", "retinular cell" ]	[ "Cornea", "Corneal facet", "cell", "cone", "Cone cell", "Rhabdome", "Optic ganglion" ]	Structure of compound eyes	extended	true	true	true
NALANDA-195	google/gemini-2.5-flash	desc	0.276	0.222	0.364	9	11	[ "haploid nucleus", "diploid nucleus", "basidiospore", "basidium", "germ tube", "primary sporidium", "sterigma", "haploid nucleus (+ strain)", "haploid nucleus (- strain)" ]	[ "Apex", "Karyogamy", "Basidiospores", "2x", "Meiosis", "Basidium", "Stalk", "young teleutospore", "mature teleutospore", "germinating teleutospore with young basidium", "germinating teleutospore with basidium and basidiospores 2+ and 2–" ]	Puccinia sp	extended	true	false	true
NALANDA-205	google/gemini-2.5-pro	blind	0.628	0.533	0.765	15	17	[ "Conidiophore", "Metula", "Conidia", "Phialide", "Ramus (Branch)", "Mycelium", "Germinating conidium", "Ascogonium", "Antheridium", "Ascus", "Ascospore", "Sterigma (Phialide)", "Conidium", "Septum", "Nucleus" ]	[ "Chains of Conidia Conidia", "Septum", "Sterigmata Metula", "Cytoplasm", "Sterigma", "Cell Wall", "Metulae", "Oil Globule", "Septal Pore", "Nucleus", "Conidiophore", "sterigmata and conidial chains arising on unbranched conidiophores", "sterigmata and conidial chains developing on branched ...	Penicillium sp. Asexual reproduction	extended	true	true	true
NALANDA-215	google/gemini-2.5-pro	no-image	0.255	0.286	0.231	7	13	[ "Glucose", "Pyruvate", "Acetyl CoA", "Carbon dioxide", "Reduced NAD", "Oxygen", "ATP" ]	[ "Protein", "Amino acids", "Carbohydrate", "Glucose", "Glucose 6-phosphate", "Fats", "Glycerol", "Fatty acid", "Glyceraldehyde phosphate", "Pyruvic acid", "Acetyl Co-A", "Citric acid cycle", "Electron transport system" ]	Botany diagram	core	true	false	true
NALANDA-52	openai/gpt-4o	blind	0.612	0.625	0.6	8	10	[ "Algal cells", "Fungal hyphae", "Asci", "Ascocarps", "Upper cortex", "Medulla", "Lower cortex", "Rhizines" ]	[ "Isidium", "Fungal Hypha", "Algal Cells", "Cortex", "Algal Zone", "Upper Cortex", "Algal Cell", "Medulla", "Cephalodium", "Algal Layer" ]	Lichens	extended	true	true	true
NALANDA-89	google/gemini-2.5-pro	no-image	0.296	0.333	0.267	12	15	[ "Apex", "Shell", "Aperture", "Optic tentacle", "Eye", "Sensory tentacle", "Mouth", "Foot", "Tail", "Pneumostome", "Mantle collar", "Genital pore" ]	[ "Gonad", "coelom", "Stomach", "Heart", "Nephridium", "Mantle", "Shell", "Radula", "Mantle cavity", "Anus", "Gill", "Visceral mass", "Mouth", "Foot", "Intestine" ]	The garden snail	extended	true	false	true
NALANDA-90	google/gemini-3.6-flash	blind	0.5	0.5	0.5	2	2	[ "Trachea", "Intercostal muscles" ]	[ "Trachea", "Pleural membrane" ]	the human respiratory system	patternc	false	true	true

BLINDSPOT — Public Sample (10 rows)

10 representative rows from Nalandadata/BLINDSPOT — no login required.

A model is shown an unlabelled scientific diagram and must name every part a leader line points to. This sample shows scored responses from the benchmark's results.jsonl file across different models and conditions.

Full dataset (6,335 scored responses, 9 models, 4 conditions, 181 items): Nalandadata/BLINDSPOT — requires access request.

Key finding

The headline result is not the leaderboard — it is the no-image control. When the picture is removed and models are only told the subject name, 8 of 9 models keep 70–100% of their score. For gpt-4o-mini, the image contributes nothing (+0.2 pts, CI contains zero).

grounded contribution = told − no-image   (percentage points)
recitation share      = no-image ÷ told   (%)

Columns

Column	Type	Description
`item_id`	string	Public item code (`NALANDA-N`)
`model`	string	Model identifier, e.g. `google/gemini-3.6-flash`
`condition`	string	`blind`, `told`, `desc`, or `no-image`
`f1`	float	Synonym-aware name-set F1 (0–1)
`precision`	float	Precision component
`recall`	float	Recall component
`n_pred`	int	Number of names the model predicted
`n_key`	int	Number of ground-truth names
`predicted_names`	list[str]	The model's label list
`key_names`	list[str]	Ground-truth anatomical names
`subject`	string	Short factual description of the diagram
`tier`	string	Item tier (`core` or `extended`)
`unsolved_core`	bool	True if no baseline model solved this item
`recitable`	bool	True if ≥1 model scored F1 ≥ 0.5 with no image
`parse_ok`	bool	Whether the model's output parsed correctly

Why no images?

All 181 benchmark items come from copyrighted textbooks and a proprietary question bank. Only scores and predicted names are released; source figures are not distributable.

For enquiries: info@nalandadata.ai · nalandadata.ai

Downloads last month: 12

Number of rows:

Total file size:

5.31 MB